AMD: Pirate Islands (R* 3** series) Speculation/Rumor Thread

Discussion in 'Architecture and Products' started by iMacmatician, Apr 10, 2014.

Tags:
  1. pharma

    Veteran Regular

    Joined:
    Mar 29, 2004
    Messages:
    3,309
    Likes Received:
    1,936
    Okay, gotcha. So it's basically marketing hype when they say "3D HBM Dram Die Size on Silicon Interposer" in Financial Analyst slide. Seems to be a lot of confusion all around!
     
  2. LordEC911

    Regular

    Joined:
    Nov 25, 2007
    Messages:
    795
    Likes Received:
    78
    Location:
    'Zona
    Well it is a "3d HBM DRAM Stack on Silicon Interposer." HBM is an IC that is 3d stacked, it just isn't stacked on top of the GPU.
     
    Grall likes this.
  3. Kaotik

    Kaotik Drunk Member
    Legend

    Joined:
    Apr 16, 2003
    Messages:
    8,706
    Likes Received:
    2,502
    Location:
    Finland
    The memories are 3D stacked, the "whole unit" aka GPU + memories is "2.5D stacked"
     
  4. lanek

    Veteran

    Joined:
    Mar 7, 2012
    Messages:
    2,469
    Likes Received:
    315
    Location:
    Switzerland
    Effectively, the 3D refer to the memory technology ( similar of the 3D Nand of Samsung ), 2.5D mean the gpu core and memory are on the same substrate interposer.. otherwise, it will mean Top PIM ( memory transistor layer over the gpu transistors ) but we will not see this before, well 2018-2020.. ( AMD have just start the research on it in 2014 )
     
  5. Newguy

    Regular Newcomer

    Joined:
    Nov 10, 2014
    Messages:
    257
    Likes Received:
    113
    Yeah that's what I was thinking. Seen claims ("claims") that clock for clock the node consumes about half the power of the current 28nm node. Of course add a bit of a clock increase for more performance and even it out with a bit of arch improvements, it'll mostly be the node accounting for it and not a major arch improvement in that regard. Unless they're really pushing the clocks and there is a large arch improvement, but I'd find that the more unlikely scenario.

    That's even assuming it reaches that, we've all seen company slides claim a lot of things over the years. Actually do we know if that improvement's over Tonga or Fiji, assuming it's new? If there are reasonable perf/w gains with Fiji and it doubles it might be quite nice.
     
  6. LordEC911

    Regular

    Joined:
    Nov 25, 2007
    Messages:
    795
    Likes Received:
    78
    Location:
    'Zona
    [​IMG]
    Looks like the VRMs might be cooled by the block.
     
    Lightman, Jwm, DieH@rd and 1 other person like this.
  7. Newguy

    Regular Newcomer

    Joined:
    Nov 10, 2014
    Messages:
    257
    Likes Received:
    113
    That is quite incredibly short.
     
    Grall likes this.
  8. Kaotik

    Kaotik Drunk Member
    Legend

    Joined:
    Apr 16, 2003
    Messages:
    8,706
    Likes Received:
    2,502
    Location:
    Finland
    But are those really legit or just some fan renderings?
     
  9. A1xLLcqAgt0qc2RyMz0y

    Veteran Regular

    Joined:
    Feb 6, 2010
    Messages:
    1,045
    Likes Received:
    355
    With HBM it should be shorter as the memory is much smaller (stacked) and placed right next to the GPU. No more need for 16 discrete GDDR5 spread all around the PCB.
     
  10. Kaotik

    Kaotik Drunk Member
    Legend

    Joined:
    Apr 16, 2003
    Messages:
    8,706
    Likes Received:
    2,502
    Location:
    Finland
    GTX 970's is actually shorter based on the image.
    GTX 970 reference PCB is 175mm while the claimed R9 390X is ~190mm
     
  11. Love_In_Rio

    Veteran

    Joined:
    Apr 21, 2004
    Messages:
    1,542
    Likes Received:
    166
    Making a comparison guided by the PCI-express length it doesn´t look so and by far.
     
  12. Kaotik

    Kaotik Drunk Member
    Legend

    Joined:
    Apr 16, 2003
    Messages:
    8,706
    Likes Received:
    2,502
    Location:
    Finland
    That's how I did it - the PCI Express -connector on the claimed R9 390X is 129 pixels and 88mm in real life, so 1 pixel is ~0,6821... mm in real life.
    The PCB length on the image is 280 pixels (I'm assuming PCB ends where the cooler shroud ends on the right side), which makes the full length about 191mm, so with some error margin ~190mm
     
  13. 3dilettante

    Legend Alpha

    Joined:
    Sep 15, 2003
    Messages:
    8,263
    Likes Received:
    3,450
    Location:
    Well within 3d
    It seems unlikely to me as well. The power limit really pushes power efficiency to the forefront, and barring running into a significant barrier elsewhere, pushing clocks enough to negate an architectural revamp would be counterproductive and more expensive than coasting.

    Unclear. AMD didn't even supply a vague, not normalized to zero, fluffy graph like it did for the CPU improvement curve.
    It is possible to infer a separation since the boxes are separate, but AMD has done much to take advantage of reasonable inferences on the part of consumers, then feign confusion that doublespeak leads to different conclusions than the ones it refused to state plainly.
     
  14. Grall

    Grall Invisible Member
    Legend

    Joined:
    Apr 14, 2002
    Messages:
    10,801
    Likes Received:
    2,174
    Location:
    La-la land
    Effin hell! If that's a genuine image...wow. Ideal card for uATX crossfire monster system!

    I was thinking, months ago, that the board ought to be considerably shorter with stacked memory, but I didn't dare believing it. ...One can only hope, eh?
     
  15. DieH@rd

    Legend Veteran

    Joined:
    Sep 20, 2006
    Messages:
    6,186
    Likes Received:
    2,115
    Very nice and compact PCB. Air cooled version will of course be more beefy.

    I can't wait for 14nm [and with it HBM2] to arrive so that I can finally get some meaningful upgrade that will make my rig VR ready.
     
  16. TomRL

    Newcomer

    Joined:
    Apr 19, 2014
    Messages:
    183
    Likes Received:
    4
    I'm new to the gpu market, are gpu's normally revealed around this time? Should I expect the opposing Nvidia reveal soon? I was considering getting 970 but if they are doing a new one :s
     
  17. Alexko

    Veteran Subscriber

    Joined:
    Aug 31, 2009
    Messages:
    4,512
    Likes Received:
    930
    There's not much of a typical seasonal pattern, it depends on when chips are ready more than anything else (and inventory). NVIDIA has just released the Titan X, and may follow up with a more traditional card based on the same GM200 GPU with 6GB of RAM and possibly faster clocks, but it's extremely unlikely that they would introduce a bigger GPU until 2016.
     
  18. Pressure

    Veteran Regular

    Joined:
    Mar 30, 2004
    Messages:
    1,389
    Likes Received:
    319
    They would work great in an updated Mac Pro with that form factor ;)

    Wouldn't be surprised if Apple were one of the first customers, as with Tonga.
     
    Grall likes this.
  19. tunafish

    Regular

    Joined:
    Aug 19, 2011
    Messages:
    571
    Likes Received:
    244
    AMD promised to it's investors that it's revealing it's new GPUs "this quarter", or before the end of June. nV has a "980 Ti", that is basically a 6GB version of the Titan X, waiting to be used as a reply to the AMD release. Depending on how good the new AMD cards are, nV could do a major pricing resuffle or they might just release the 980Ti at some ridiculously high price point.
     
  20. Razor1

    Veteran

    Joined:
    Jul 24, 2004
    Messages:
    4,232
    Likes Received:
    749
    Location:
    NY, NY
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...